CN101388052A - Method for the operation of a multiprocessor system in conjunction with a medical imaging system - Google Patents
Method for the operation of a multiprocessor system in conjunction with a medical imaging system Download PDFInfo
- Publication number
- CN101388052A CN101388052A CNA2008101686198A CN200810168619A CN101388052A CN 101388052 A CN101388052 A CN 101388052A CN A2008101686198 A CNA2008101686198 A CN A2008101686198A CN 200810168619 A CN200810168619 A CN 200810168619A CN 101388052 A CN101388052 A CN 101388052A
- Authority
- CN
- China
- Prior art keywords
- data
- processing unit
- aforementioned
- computing
- algo1
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5066—Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/54—Interprogram communication
- G06F9/544—Buffers; Shared memory; Pipes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/50—Indexing scheme relating to G06F9/50
- G06F2209/5017—Task decomposition
Abstract
The invention relates to a method for operating a multiprocessor system, especially in conjunction with a medical imaging system. The invention also relates to a medical imaging device which is designed to perform this method. The multiprocessor system in this case has at least two processing units (PZ1, PZ2, PZ3), at least one control unit (K) and operations (Algo1, Algo2, Algo3) which can be allocated to the processing units (PZ1, PZ2, PZ3). Data provided from an input are processed by the processing units (PZ1, PZ2, PZ3) and made available at an output. The at least one control unit (K) enhances the named data with control data (A1, A2, A3), which defines an allocation of the data to the respective operations (Algo1, Algo2, Algo3) for the purposes of processing.
Description
Technical field
The present invention relates to a kind of method that is used for particularly the operation multicomputer system relevant with medical image system.The invention still further relates to a kind of in addition for implementing the medical image system that this method is constructed.
Background technology
Be used for the typical X ray equipment of insertion type angiography, producing the time sequencing of radioscopic image.All the time carry out the processing of single image according to identical mode, wherein, the speed of handling is proposed certain requirement.For handling the algorithm that image is used for image improvement.These algorithms are realized with the program mode (PM) of the conversion of embodiment image information.Typically only in single computing unit (CPU, DSP, FPGA, ASIC etc.), can not finish processing, but must divide a plurality of steps on a plurality of computing units, to carry out the intensive of radioscopic image.Usually the pipeline organization of the step that resolves into single order will be all handled in selection for this reason.Step can be distributed on a plurality of processing units then, because treatment step is independently of one another.At this, for each treatment step is equipped with the parameter group that step is dealt with in a control separately.For example, the dynamic (dynamical) treatment step needs of gray-scale value " Windowing " that for example are used for rebuilding the parameter group of forming by " center (Center), width (Width) ".This information is transferred to treatment step by rights.
In widely used pipeline organization, on these data that must handle in chronological order, carry out " stream line operation " of treatment step.When the stream line operation of data, on discrete time point, give a processing unit (" process "), the first of this processing unit computational algorithm exclusive disjunction (" program ") with newly arrived data allocations.After this calculates end, intermediate result is sent to next processing unit, this processing unit applies to this data with the next step of algorithm then.So repeatedly repeatedly, until carrying out institute in steps and obtain the result.At this, the quantity of performed treatment step is called " degree of depth " of streamline.In other words, the realization of pipeline processes be with data to be processed be considered as a kind of " river " that " flows " by single treatment step.
Fig. 1 illustrates a kind of like this example of realizing pipeline processes.Fig. 1 illustrates data source Q and meeting point S, and wherein, data transmit to meeting point S from data source Q.A control module K who is also referred to as controlled entity (Kontrollinstanz) is shown in addition, its control data from a processor for example PZ1 to for example transmission of PZ2 of next processor.In illustrating respectively, this adorns a plurality of processor P Z1 to PZ3 of a kind of algorithm ALGO1, ALGO2, ALGO3.External interface adopts IN to be labeled as input end and adopts OUT to be labeled as output terminal.In single treatment step, provide parameter group A1, A2, A3 by independent mechanism by control module K control.For example, when pipeline processes is moved, can influence so-called " Windowing " during handling by the parameter group new or that change of handling level is provided.As control information be stored in the control data parameter group will with (effectively) synchronization of data streams.Reach thus, since definite for example n the parameter variation that data set is determined.This synchronous requirement in the pipeline architecture is to the specific knowledge of the topological structure of entire process system up to now.Therefore to each time point all to provide data set number just in time the information of residing processing unit to higher level's control module.In this manner, can on correct time point, indicate the parameter group that changes.Another shortcoming of this type realization is, can not easily insert additional treatment step.Control module needs the information of the relevant processing duration that changes for this reason.Bring the granularity (Granularit of deal with data similarly
T) variation is difficult.For example, if a treatment step is carried out based on the algorithm of going, but next treatment step is handled the data of being made up of three row (for example carrying out three cores (Dreier-Kernels)), the data buffering in streamline, also will consider the delay in the parameter group application so.These difficulties are by following solution at present, be them or be left in the basket and disregard (parameter group of change influences next data immediately), by higher level's information (frame number for example, Frame-Number) carry out the application that rises from next suitable data structure (different granularities, for example pixel-wise or line mode).
Summary of the invention
The technical problem to be solved in the present invention is to overcome above-mentioned shortcoming.
Main aspect of the present invention is a kind of particularly method of the operation multicomputer system of relevant medical image system that is used for.
In this system, have at least two processing units, at least one control module and can distribute to the computing of processing unit.The data of importing from input end by processing unit processes also can offer output terminal.At least one control module utilizes control data to strengthen alleged data, and this control data is for the transmission of the purpose specified data of handling to each computing.
Preferably distribute each computing to processing unit by at least one control module.Suitably, can by predetermined order determine or control data to the transmission of each computing.
Can be sequentially or deal with data concurrently.So at least, another processing unit or governable processing unit can be used to the processing of data, and at least one computing is assigned to this processing unit.
Suitably, the data of being imported by input end comprise valid data and control data, and wherein, control data preferably can be arranged in one so-called " header (Header) ".
During handling or afterwards, mate control data by at least one computing of distributing to processing unit.Thus can specified data to each computing again or the transmission that repeats.
Preferably data are transferred to each computing and processing chronologically.
Can consider the logic module of multicore computing unit or cluster computer, many DSP configuration, Cell processor, stream handle or freely programmable as multicomputer system.
Another embodiment of the invention is, for the processing unit of its distributive operation by at least one shared storage unit swap data.
Also can consider, except or replace by a shared storage unit, carry out swap data by at least one shared connection network.
Can connect network by at least one and connect connection between the processing unit statically or dynamically.
Exchanges data also can connect network by at least one and carry out in the mode of packet.
The transmission of data can be carried out by incident control or by sequential control.
Also can use one and be also referred to as the processing storehouse (Bearbeitungspool) of " workman storehouse (worker pools) ".At this, so-called scheduler program control is to the data transmission of processing unit.
Another aspect of the present invention is a kind of medical imaging apparatus, and this device is configured to implementation basis method of the present invention and embodiment thereof.Can have connection network between the processing unit suitably by a kind of device, wherein can be statically and/or dynamically connect this connection.
The present invention has following advantage:
Reach following advantage by strengthening valid data by whole control datas:
-control data and valid data are being handled any locational simple synchronization of streamline,
-utilize also the additional treatment step in pipeline processes can the extension process system,
-set up the possibility (handling network, division and combination once more) of non-linear topological structure,
No longer specific just in time treatment step that turns to of calculating in-" processing unit ",
-compare and can arrange the processing time more neatly with the tupe of traditional strict sequential order,
-can realize algorithm (just by convergence and control computing time, the quantity that cycles through is not fixing from the beginning) by the control of convergence standard,
-utilize the possibility of unspecific processing unit,
The extended capability of-modular, the words of higher if desired data rate.That is to say, under the situation of reservation operations lag period, use additional processing unit,
-utilize the possibility of specific and general process source,
-physical topological structure flexibly, star preferably, wherein, (being preferably linear) topological structure that can configuration logic.
Description of drawings
The present invention is described in detail by one or more embodiment by accompanying drawing below.Wherein:
Fig. 1 illustrates the typical pipeline architecture according to prior art that this paper beginning is mentioned;
Fig. 2 is exemplary to be illustrated according to the enhancing of active traffic of the present invention by control data;
Fig. 3 illustrate data processing with a shared storage according to embodiments of the present invention; And
Fig. 4 illustrate by connect network carry out data processing according to embodiments of the present invention.
In Fig. 2,3 and 4, corresponding Reference numeral among identical to a certain extent unit use and Fig. 1.
Embodiment
Fig. 2 illustrates a kind of foundation pipeline architecture of the present invention, wherein, with parameter group P as control data for example A1, A2 or A3 add in the valid data, and utilize data stream " to send " next processing unit PZ2 or PZ3 from a processing unit PZ1 or PZ2 (processor).Only represent among the figure, for example mate control data A1, A2, A3 and be delivered to next treatment step after Algo1, Aglo2, the Aglo3 at each treatment step by a arrow that separates on data stream arrow next door.
Just as previously mentioned, not only embody necessary input data or valid data (pixel, gray-scale value) in the data structure that is used for pipeline processes, but also will be in independent and additional structural unit be the processing level input parameter group of streamline back.Strengthen the input data in special position (for example in so-called header) by having the control module that is used for the indication further handled thus, and the processing level of back can with under the situation that control module continues to be connected independently not carry out its each treatment step.All processing is undertaken by data manipulation fully and is therefore asynchronous to a certain extent.In other words, handle on the level, not only receive intrinsic valid data as the input data set at each, and the additional control data that receives.Handle level and can from these control datas, extract the required parameter group of this step, and arrive input data or valid data, and produce output data thus according to algorithm application.Then according to being further processed like that of by the agency of.
The advantage of this method is, fully phase out control module and single processing level complexity and easily make mistakes synchronously.Parameter group is put with the valid data direct correlation at any time and can be used for each treatment step.
Another advantage is, can be summarised as " unit system structure " from the pipeline architecture of strictness.In pipeline architecture, by each processing unit only with a kind of algorithm application to data, handle level according to this algorithm by specialization and fixed time sequence work to predesignate.In now possible unit system structure, for example processing unit can decide according to the standard of the processing duration that is allowed, implements a plurality of steps (also can be algorithms of different).In this manner, realize for example iterative algorithm more simply.
If data set allows the short processing time, for example owing to very rapidly satisfy the condition of convergence, then this treatment step also can more promptly finish.If same data set needs the processing time slightly longer in next processing level, so present next processing level also can use longer computing time.Replace the tupe of main up to now strict sequential order and form a kind of nonsynchronous tupe on the whole.
Particularly utilize the parametrization of treatment step that to implement the long duration of these variations according to the solution of the present invention simply.Therefore realize the algorithm of bigger quantity than pipeline processes chronologically based on the processing of unit.At this, can exist than the more algorithm of processor, wherein, can be a more algorithm of processor distribution.
In addition, on the meaning that simplify to realize, consider a kind of logical process chain, wherein, also can consider other topological structure (for example branch, converge etc.).
According to the unitized construction of valid data and control data, also can use a kind of foundation " blackboard " model with at least one shared storage SP-as shown in Figure 3-disposal route.At this, data are in a shared storer SP " disclosing ".Unappropriated processing unit for example PZ1 is made a response to these disclosed tasks now, method is that it receives data, by appended control data next treatment step is applied to valid data, and will stores shared storage again into the result of the additional marking institute mark of the present treatment step of implementing.This process can be controlled based on incident and carry out.This point also can realize by the processing storehouse that is also referred to as " workman storehouse ".At this, so-called scheduler program control is assigned to one of processing unit with task.
A major advantage of this model is that algorithm and processor (or processing unit) are separated from each other now.Therefore not only can sequential processes, and can parallel processing.Can pay attention to when the data processing as expansion for processor or other processors of domination, method is that algorithm distributes different processors (for example DSP, unit core, multicore, clustered processors, stream handle or FPGA) according to the type (capacity, ability) of processor.With reference to figure 4 time, for example can consider by following processor P Z3 closely the processor with algorithm Algo4 for example PZ4 replenish this synoptic diagram as expansion.
Another kind of transaction module can be by following realization, promptly data except or replacement be stored in the shared storage-as shown in Figure 4-transmit data by a shared connection network VN.This point can be considered as a kind of embodiment of shared storage (" blackboard " model), wherein only to each shared in pairs output and input data modeling.The dynamically configuration property again of " packet switching network " by using classification, almost can reach with by the identical dirigibility of above-mentioned " blackboard " model.With the switching of processing unit can be by the realization of sending of so-called multileaving grouping.In most cases, determine in advance and by the desired topological structure of this network misconnection (normally linear).
Claims (18)
1. method that is used for particularly the operation multicomputer system relevant with medical image system, wherein there are at least two processing unit (PZ1, PZ2, PZ3), at least one control module (K) and can distribute to described processing unit (PZ1, PZ2, PZ3) computing (Algo1, Algo2, Algo3), wherein, by described processing unit (PZ1, PZ2, PZ3) handle the data of importing from input end (IN) and also can offer output terminal (OUT), it is characterized in that, described at least one control module (K) utilizes control data (A1, A2, A3) strengthen alleged data, described control data determines that for the purpose of handling described data are to described each computing (Algo1, Algo2, Algo3) transmission.
2. by each described method in the aforementioned claim, it is characterized in that described data are determined by predetermined order to the transmission of described each computing (Algo1, Algo2, Algo3).
3. by each described method in the aforementioned claim, it is characterized in that described each computing (Algo1, Algo2, Algo3) is assigned to described processing unit (PZ1, PZ2, PZ3) by described at least one control module (K).
4. by each described method in the aforementioned claim, it is characterized in that described data make to the transmission of described each computing (Algo1, Algo2, Algo3) can be sequentially and/or processing said data concurrently.
5. by each described method in the aforementioned claim, it is characterized in that, for processing said data is dynamically replenished another processing unit at least, for this another processing unit can distribute at least one computing.
6. by each described method in the aforementioned claim, it is characterized in that the data of being imported by input end comprise the valid data that utilize alleged control data (A1, A2, A3) to strengthen.
7. by each described method in the aforementioned claim, it is characterized in that, during handling or afterwards, mate described control data (A1, A2, A3) like this by at least one computing (Algo1, Algo2, Algo3) of distributing to processing unit, make these control datas determine described data to described each computing (Algo1, Algo2, Algo3) again or repeat the transmission.
8. by each described method in the aforementioned claim, it is characterized in that, transmit described data chronologically to described each computing (Algo1, Algo2, Algo3) and processing.
9. by each described method in the aforementioned claim, it is characterized in that, utilize the logic module of multicore computing unit or cluster computer, many DSP configuration, Cell processor, stream handle or freely programmable to operate multicomputer system.
10. by each described method in the aforementioned claim, it is characterized in that, for the processing unit (PZ1, PZ2, PZ3) of its distributive operation (Algo1, Algo2, Algo3) exchanges described data by at least one shared storage unit (SP).
11. by each described method in the aforementioned claim, it is characterized in that, for the processing unit (PZ1, PZ2, PZ3) of its distributive operation (Algo1, Algo2, Algo3) exchanges described data by at least one shared connection network (VN).
12., it is characterized in that the static state between the described processing unit (PZ1, PZ2, PZ3) connects by described at least one connection network (VN) to be connected by each described method in the aforementioned claim.
13., it is characterized in that the connection between the described processing unit (PZ1, PZ2, PZ3) is dynamically connected by described at least one connection network (VN) by each described method among the claim 1-11.
14. by each described method in the aforementioned claim, it is characterized in that, can exchange described data in the mode of packet by described at least one connection network (VN).
15., it is characterized in that the transmission of described data is undertaken by incident control by each described method among the aforementioned claim 10-14.
16. by each described method among the aforementioned claim 10-14, it is characterized in that, carry out of the transmission of described data to the processing unit that belongs to so-called processing storehouse by so-called scheduler program.
17. a medical imaging apparatus, it is configured to implement by each described method at least in the aforementioned claim.
18. by the described device of claim 17, it is characterized in that this device has a connection network between the processing unit, wherein can be statically or dynamically connect described connection.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102007034684A DE102007034684A1 (en) | 2007-07-25 | 2007-07-25 | Method for operating a multiprocessor system, in particular in connection with a medical imaging system |
| DE102007034684.2 | 2007-07-25 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101388052A true CN101388052A (en) | 2009-03-18 |
Family
ID=40157182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2008101686198A Pending CN101388052A (en) | 2007-07-25 | 2008-07-25 | Method for the operation of a multiprocessor system in conjunction with a medical imaging system |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US20090031119A1 (en) |
| CN (1) | CN101388052A (en) |
| DE (2) | DE202007018934U1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109634691A (en) * | 2017-10-09 | 2019-04-16 | 罗伯特·博世有限公司 | Computing unit and operation method to this |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7700469B2 (en) | 2008-02-26 | 2010-04-20 | Micron Technology, Inc. | Methods of forming semiconductor constructions |
| GB2553597A (en) * | 2016-09-07 | 2018-03-14 | Cisco Tech Inc | Multimedia processing in IP networks |
| CN108874548B (en) * | 2018-07-11 | 2021-04-02 | 深圳市东微智能科技股份有限公司 | Data processing scheduling method and device, computer equipment and data processing system |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS59132070A (en) * | 1983-01-18 | 1984-07-30 | Mitsubishi Electric Corp | Data processing device for array operation |
| US5522083A (en) * | 1989-11-17 | 1996-05-28 | Texas Instruments Incorporated | Reconfigurable multi-processor operating in SIMD mode with one processor fetching instructions for use by remaining processors |
| US6118452A (en) * | 1997-08-05 | 2000-09-12 | Hewlett-Packard Company | Fragment visibility pretest system and methodology for improved performance of a graphics system |
| US5915123A (en) * | 1997-10-31 | 1999-06-22 | Silicon Spice | Method and apparatus for controlling configuration memory contexts of processing elements in a network of multiple context processing elements |
| US6092174A (en) * | 1998-06-01 | 2000-07-18 | Context, Inc. | Dynamically reconfigurable distributed integrated circuit processor and method |
| US7167148B2 (en) * | 2003-08-25 | 2007-01-23 | Texas Instruments Incorporated | Data processing methods and apparatus in digital display systems |
-
2007
- 2007-07-25 DE DE202007018934U patent/DE202007018934U1/en not_active Expired - Lifetime
- 2007-07-25 DE DE102007034684A patent/DE102007034684A1/en not_active Ceased
-
2008
- 2008-07-22 US US12/220,130 patent/US20090031119A1/en not_active Abandoned
- 2008-07-25 CN CNA2008101686198A patent/CN101388052A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109634691A (en) * | 2017-10-09 | 2019-04-16 | 罗伯特·博世有限公司 | Computing unit and operation method to this |
Also Published As
| Publication number | Publication date |
|---|---|
| DE102007034684A1 (en) | 2009-01-29 |
| US20090031119A1 (en) | 2009-01-29 |
| DE202007018934U1 (en) | 2010-01-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104657220B (en) | Scheduling model and method based on deadline and expense restriction in mixed cloud | |
| CN103229146B (en) | Computer cluster for handling calculating task is arranged and its operating method | |
| CN104375882B (en) | The multistage nested data being matched with high-performance computer structure drives method of calculation | |
| CN103699446A (en) | Quantum-behaved particle swarm optimization (QPSO) algorithm based multi-objective dynamic workflow scheduling method | |
| CN101978659A (en) | Express virtual channels in a packet switched on-chip interconnection network | |
| CN108415771A (en) | Multi-chip distributed parallel computing acceleration system | |
| CN103731372A (en) | Resource supply method for service supplier under hybrid cloud environment | |
| CN108845874A (en) | The dynamic allocation method and server of resource | |
| CN105468546B (en) | Data processing apparatus and method for interconnection circuit | |
| CN106502782A (en) | Heterogeneous computing system and its method | |
| CN101388052A (en) | Method for the operation of a multiprocessor system in conjunction with a medical imaging system | |
| US20120204183A1 (en) | Associative distribution units for a high flowrate synchronizer/schedule | |
| CN109840877A (en) | A kind of graphics processor and its resource regulating method, device | |
| CN105786447A (en) | Method and apparatus for processing data by server and server | |
| CN109213588A (en) | A kind of cloud data center Batch Arrival task allocation apparatus, system and method | |
| CN106294445A (en) | The method and device stored based on the data across machine room Hadoop cluster | |
| Dong et al. | Slardar: Scheduling information incomplete inter-datacenter deadline-aware coflows with a decentralized framework | |
| KR20210137472A (en) | Pipeline arithmetic unit, programmable logic controller, and execution method of pipelined processing | |
| US20230185577A1 (en) | Communication in a Computer Having Multiple Processors | |
| US11940940B2 (en) | External exchange connectivity | |
| Jaggi et al. | Periodic inventory model with reduced setup cost under service level constraint | |
| US11625357B2 (en) | Control of data transfer between processors | |
| US11726937B2 (en) | Control of data sending from a multi-processor device | |
| CN107301085A (en) | A kind of cloud platform method for allocating tasks based on queue | |
| US20220374951A1 (en) | Charging for the use of resources in a distributed network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20090318 |